A walkthrough of using Grafana Alloy's eBPF-based profiler and Pyroscope to analyze C++ performance bottlenecks in TON blockchain validator code. Using flame graphs and diffs, the team identified hotspots like SHA256 hashing (14% of execution time) and Ed25519 signature verification. Key optimizations included swapping crypto libraries (1.5–3.5% gains), consolidating SHA256 feed calls (20% speedup on DataCell::create), and replacing std::map with std::unordered_set (10% speedup). The post also highlights how eBPF profiling eliminates the need for custom instrumentation code, contrasting it with the manual tracing and memory profilers contestants built themselves.
Sort: